Methodology

This page explains exactly how a Worth or Skip review reaches its conclusion, and where the limits of what we can actually confirm sit. If a review contradicts what’s written here, the review is wrong — tell us through the Contact page.

The conclusion, in plain words

Every review ends in a written conclusion: we either recommend the product, with the condition attached (“worth buying if…”), or we don’t, with the specific reason named. There is no star rating, no score out of ten, and no seal of approval. If a product only makes sense for a specific kind of buyer, that lives in the “Who it’s not for” section — nuance belongs in the text, never in a badge.

The recommendation measures business conduct, not teaching quality or technical excellence: “we recommend it” is not the same claim as “this is the best in its category.” It means the seller delivers what it promises and doesn’t burn the buyer.

What makes us say “don’t buy it”

We do not recommend a product if any of these is true:

1. Regulatory action — a ban, fine, court order, or consumer-protection ruling against the product or the seller.
2. A complaint pattern about the core promise, measured by rate, not raw count. We count complaints naming the product on public complaint platforms and look at the resolution rate the platform itself calculates, not just the headline number. A rate needs a minimum sample — roughly 10 named complaints — before it’s statistically trustworthy; below that, a bad-looking number doesn’t auto-fail the product, it becomes a declared gap instead (more on this below). One distinction matters here: complaints about delivery, refunds, billing, or cancellation get counted per company — it’s the same fulfillment operation behind every product that company sells, and slicing the count by individual product would hide how big the real problem is. Complaints about an adverse effect or the product’s own composition get counted per product — that doesn’t transfer between a company’s different products.
3. The sales page describes something materially different from what checkout delivers — and “materially” is the actual bar, not any difference at all. It has to change what the buyer receives, what they pay, or the risk they’re taking on. A promotional checkout price differing from a page’s list price doesn’t clear that bar on its own.
4. The seller has an unresolved track record on the same promise — a recurring complaint pattern the company has never fixed.
5. The offer isn’t verifiable. No traceable checkout exists for the product — verified, not assumed; this is not “the page is hard to find,” it’s “there is no way to buy this in a way that can be tracked.” This is never about whether the seller is a registered company. A sole proprietor with a readable page and a visible price is a verifiable seller, same as most legitimate independent creators.

What makes us recommend it

We recommend a product if none of the triggers above fires, and at least one of these is verified positively — a real “yes,” checked by us, not just the absence of red flags:

• Checkout confirmed — we personally followed the buy link through to the end, and it delivers exactly what the page promises, with no gap in product, seller, or price.
• Seller identifiable — a traceable person or company, with a page, a name, and a checkable public history. We never require formal business registration: an individual seller with a locatable offer and a visible price is verifiable, and the absence of a formal company never counts against a product — alone or stacked with anything else.
• Guarantee confirmed — the refund window and mechanism verified, ideally backed by a real report from someone who claimed it and got paid.
• Real, dated reviews — review volume counted and dated, with a source. Not “the rating is high” — it’s “of 40 reports read, 12 say they got X; 4 say they didn’t.” A small, new product with a handful of real, dated reviews clears the bar here; what matters is that the reviews exist and trace back to a source, not that there are hundreds of them.

Benefit of the doubt: when one of these four genuinely can’t be checked for lack of available data — not for lack of looking — it doesn’t count against the product, and moves into the “what we couldn’t verify” note instead. This applies most often to outcome reports: someone who finishes a $30 course rarely writes publicly about it, and failing a product for that silence would punish every legitimate small product by default. The line that stops this from becoming a loophole: the benefit only applies where we actually looked and found nothing, with a source and a date on record. A blank field earns nothing — that’s incomplete work, not a legitimate gap.

Above $100, at least one of the four needs to be an explicit, counted, dated volume signal — the absence of red flags alone isn’t enough at that price point. The damage to a reader scales with price, so the bar for proof scales with it too. When that doesn’t close, the review still publishes — recommended, with the gap stated plainly in the body, never silently clean.

What counts as an independent source

Independence means a different collection mechanism and a different incentive — not just a different URL. Five families of source: the sales platform itself (real buyer, but incentive aligned with the seller); a complaint channel (dissatisfied buyer, self-selects toward the negative); the seller’s or an affiliate’s own channel (favorable audience by nature); a forum or social post with no commercial relationship; and an institutional source (regulator, courts, press). We count at most one source per family, and the minimum set of three needs at least one adversarial source and one non-adversarial one. Three review sites repeating the same number from the seller’s own page is one source wearing three outfits, not three sources.

Two tie-breaking rules

• If the investigation is genuinely incomplete, the review doesn’t publish — neither recommended nor not. That’s different from “went looking for all four positives and none of them closed”: if no trigger fired but none of the four positives closed either, the result is a recommendation with the gap declared, not an unpublished review. “Doesn’t publish” is reserved for when there’s still real looking left to do — a blank field, no source, no date.
• When the complaint count lands right at the reliability threshold, the result is a recommendation with the gap declared, never a clean one. If the number of named complaints sits close to the minimum sample (say, 8 or 9 real complaints when the floor is 10), that doesn’t disappear — the number shows up in the body, the review states the evidence landed right at the edge, and the reader can redo the math themselves before deciding. A close call never becomes silent approval.

Every number shows its source, in the text

Every count, price, or claim in a review carries its source and the date it was checked, visible to the reader, in the sentence or the table row where the number appears. “12 of 40 reports mention X, read on [platform] on [date]” is accepted; “the gap is small” is not. Numeric scores are banned — there’s no 7.4 out of 10 here, because the only subjective thing in an auditable system would be where the weighting comes from.

Beyond the sources, every review states, in its own body:

1. What would change our mind — what specific evidence would flip this conclusion. It makes the conclusion falsifiable instead of an opinion, and gives a seller an exact path to respond.
2. Declared counter-evidence — what we found that weakens our own conclusion. If we found nothing, we say where we looked.
3. What we couldn’t verify — explicit gaps: a claim we couldn’t source, a review count the platform doesn’t expose.
4. Corrections, dated, in the page — a review that materially changes says so in its own body, with the date and the reason, including when the reason is “we miscounted.”

One person testing is an opinion. Forty reports with a date is a pattern.

Buying a product and trying it once is exactly that: one person, one unit, one context, one moment — an anecdote with a receipt attached. The market treats that as the gold standard because it looks honest, not because it’s rigorous. Reading forty real buyers, each with an outcome and a date, is stronger than one person’s single use — it’s what actually shows a pattern: how many hit the same problem, whether a refund gets honored when asked, whether support answers.

So most reviews here don’t start from a purchase — they start from reading what dozens of buyers reported, cross-checked against the sales page, the price, the refund policy, and the seller’s public track record. That doesn’t loosen the rule against claiming a test that didn’t happen: every review states plainly which of three situations it’s in, no exceptions.

• Evidence review (the most common method here). We read what dozens of buyers reported publicly, the sales page, the documentation, the price, the refund policy, and the seller’s public track record, and applied the criteria above. A disclosure states this near the top of the review.
• Hands-on. We bought or received the product and used it for the period described in the review. We say how long, and whether it was purchased or provided free — this answers a different question (what using it is like), not a better one.
• Comparison. A spec, price, and terms comparison between competing products, with the source for every number.

Where the numbers come from

Price, specs, and commercial terms come from the primary source — the maker’s or seller’s own page — and carry the date they were checked. Prices change weekly; if the review is old, the number is old, and the date is there so you know that.

We don’t use a number we can’t trace to its origin. “Studies show” without the study doesn’t make it in.

Why we publish negative reviews

Because a review site where everything is worth buying isn’t a review site — it’s a catalog.

The negative review is what gives the positive one meaning. If you never see us advise against a purchase, you have no way of knowing whether our recommendation means anything. Publishing “don’t buy this” on a product with a live, paying affiliate program costs real money in the short term — which is exactly why it’s the strongest proof the conclusion isn’t for sale. And a product we don’t recommend carries no buy link at all: it points you at a reviewed alternative instead.

Coverage policy

We cover products with enough public evidence to apply the full criteria in either direction. A product without that base doesn’t become a review here. There’s no published “not sure” — there’s the absence of a review, until the evidence shows up.

Conflict of interest

This site earns affiliate commission. That’s disclosed at the top of every page carrying an affiliate link, before the text starts — always the same fact: the price you pay is the same with or without the link, the commission comes out of the seller’s margin. A review with no affiliate link states that too, in reverse: no commission is involved at all.

What we don’t do: accept payment to write a review, accept pre-written copy from a manufacturer, publish an exact commission amount or percentage, or trade a conclusion for a bigger commission. If a seller ever offers that, the offer becomes the story.

Corrections and updates

A review with an error gets corrected, and the correction stays visible in the review’s own body, dated — we don’t delete and pretend it didn’t happen. A genuinely revised review gets an updated date next to the publish date, always both together. If you spotted something wrong, email roliveira.sagan@gmail.com with your source.

Changelog for this page

Criteria change when a real case exposes a hole in them — and when they do, this page changes with them. Demanding transparency from a product while hiding our own wouldn’t make sense.

September 11, 2026 — the stamp retired: reviews no longer end in a WORTH IT / SKIP stamp, and the scorecard table that sat under it is gone. The criteria on this page did not change — what changed is where the conclusion lives: in the review’s own written conclusion, with the reason on the same line, and in the body’s sourced numbers, instead of in a badge. Two things drove this: nuance kept ending up crushed into a binary stamp that the criteria themselves never treated as binary (“recommended, if…” was always the real shape of the conclusion); and a stamp is the format of an authority, while this site’s actual claim is the format of a method. The editorial voice also moved from “I” to “we” — the byline stays Roberto Oliveira, the standard is the house’s. Nothing about the affiliate rules changed: a product we don’t recommend still carries no buy link.

August 1, 2026, a change of posture: we stopped treating evidence review as an apology and started stating it as a method. Buying and trying one unit is also n=1 — one person, one context, one moment — and the market treats that as the gold standard because it looks honest, not because it’s rigorous. Reading dozens of real buyers, each with an outcome and a date, is the stronger method for the question that actually decides a purchase: does this seller deliver consistently? This doesn’t loosen the ban on claiming a test that didn’t happen — that stays absolute.

August 1, 2026: four corrections, all triggered by real cases analyzed the same day. First: complaints about delivery, refunds, billing, or cancellation are now counted per company rather than per individual product — it’s the same sales operation behind every product that company sells, and counting product-by-product hid how big the real problem was. Complaints about an adverse effect or the product’s own composition are still counted per product. Second: the minimum complaint count (10) stopped being, on its own, a reason to fail a product — it became a minimum sample size for the resolution rate to be statistically trustworthy. A flat complaint floor punished scale and protected low-volume sellers; now the rate decides, not the raw number. Third: a gap between what the sales page promises and what checkout delivers only fails a product when it changes what the buyer receives, pays, or risks. Fourth: when the complaint count lands right at the reliability threshold, the result stopped being a clean pass — it’s now a pass with the gap declared, number included.

July 31, 2026: closed the unverifiable-offer criterion (no traceable checkout for the product), explicitly decoupled from any business-registration requirement. Added two conditions to a positive conclusion: at least one criterion verified positively (not just the absence of negative signals), and an explicit volume signal required above $100. Formalized “benefit of the doubt”: a criterion with no available data — verified, with source and date, not skipped — doesn’t count against the product.

July 30, 2026: original publication of this page.